[Top] [All Lists]

Re: Pid: 8345, comm: rsync Not tainted #1

To: Stefan Priebe - Profihost AG <s.priebe@xxxxxxxxxxxx>
Subject: Re: Pid: 8345, comm: rsync Not tainted #1
From: Emmanuel Florac <eflorac@xxxxxxxxxxxxxx>
Date: Wed, 27 Oct 2010 12:06:28 +0200
Cc: Michael Monnerie <michael.monnerie@xxxxxxxxxxxxxxxxxxx>, xfs@xxxxxxxxxxx
In-reply-to: <4CC7CDF6.3080405@xxxxxxxxxxxx>
Organization: Intellique
References: <4CC67450.9020602@xxxxxxxxxxxx> <4CC6B56B.7020007@xxxxxxxxxxxx> <20101026132505.6af29d5d@xxxxxxxxxxxxxxxxxxxx> <201010270124.44846@xxxxxx> <20101027080454.31a59a3c@xxxxxxxxxxxxxx> <4CC7CDF6.3080405@xxxxxxxxxxxx>
Le Wed, 27 Oct 2010 09:00:06 +0200
Stefan Priebe - Profihost AG <s.priebe@xxxxxxxxxxxx> écrivait:

> So what values do you recommand in general for iSCSI on Serverside
> and which values on iSCSI client side?

Depends upon the hardware. nr_requests should be 512 for 3Ware cards,
256 to 1024 for other controllers (always more than the default 128

Same rule goes for read-ahead; however many RAID controllers (Areca,
Adaptec) "cheat" and do read-ahead optimisation by themselves (which
may harm random access). The default read-ahead value (256) was OK for
obsolete ATA drives with a few KB of cache; nowadays all drives have 32
or 64 MB, and controllers 512 MB to several GB, so more read-ahead can
do no harm but fill the cache up. I often go up to 65536 or 131072
sectors (for 24 or 48 drives arrays). Keep in mind that this is
sequentail-raid optimisation; random access (such as VM work) may give
better result with lower read-ahead values. Benchmarking your
application is king here.

Just for information : I currently run some servers with kernel (march 2008) and 3Ware firmware, so upgrading your
2.6.23 system should be perfectly safe from a driver compatibility
point of view.

Emmanuel Florac     |   Direction technique
                    |   Intellique
                    |   <eflorac@xxxxxxxxxxxxxx>
                    |   +33 1 78 94 84 02

<Prev in Thread] Current Thread [Next in Thread>